PURPOSE: To investigate patient set-up, tumor movement and shrinkage during 3D conformal radiotherapy for non-small cell lung cancer. MATERIALS AND METHODS: In 97 patients, electronic portal images (EPIs) were acquired and corrected for set-up using an off-line correction protocol based on a shrinking action level. For 25 selected patients, the orthogonal EPIs (taken at random points in the breathing cycle) throughout the 6-7 week course of treatment were assessed to establish the tumor position in each image using both an overlay and a delineation technique. The range of movement in each direction was calculated. The position of the tumor in the digitally reconstructed radiograph (DRR) was compared to the average position of the lesion in the EPIs. In addition, tumor shrinkage was assessed. RESULTS: The mean overall set-up errors after correction were 0, 0.6 and 0.2 mm in the x (left-right), y (cranial-caudal) and z (anterior-posterior) directions, respectively. After correction, the standard deviations (SDs) of systematic errors were 1.4, 1.5 and 1.3 mm and the SDs of random errors were 2.9, 3.1 and 2.0 mm in the x-, y- and z-directions, respectively. Without correction, 41% of patients had a set-up error of more than 5 mm vector length, but with the set-up correction protocol this percentage was reduced to 1%. The mean amplitude of tumor motion was 7.3 (SD 2.7), 12.5 (SD 7.3) and 9.4 mm (SD 5.2) in the x-, y- and z-directions, respectively. Tumor motion was greatest in the y-direction and in particular for lower lobe tumors. In 40% of the patients, the projected area of the tumor regressed by more than 20% during treatment in at least one projection. In 16 patients it was possible to define the position of the center of the tumor in the DRR. There was a mean difference of 6 mm vector length between the tumor position in the DRR and the average position in the portal images. CONCLUSIONS: The application of the correction protocol resulted in a significant improvement in the set-up accuracy. There was wide variation in the observed tumor motion with more movement of lower lobe lesions. Tumor shrinkage was observed. The position of the tumor on the planning CT scan did not always coincide with the average position as measured during treatment.